Cognition bought Poke. What it means if you're buying an AI assistant

On 23 July 2026, Cognition — the AI lab known for the Devin coding agent — announced it had acquired The Interaction Company, the startup behind Poke, the assistant you text in iMessage. Reported deal value was in the low nine figures. Cognition said Poke keeps running and picks up its models and infrastructure, and that Poke's conversational style would be brought into Devin.

We compete with Poke, so read this accordingly. We are not going to tell you the product is doomed, because we have no reason to think it is and the claim would not survive contact with reality. What the deal does is change which questions are worth asking, and those questions apply to us as much as to them.

What the deal actually signals

Interface is being priced as a moat. Cognition did not buy a model or a dataset. It bought an interaction style — the thing that makes texting Poke feel like texting a person rather than filing a request. That is a striking thing to spend nine figures on, and it says something true: the hard part of an assistant stopped being capability a while ago. Everyone's underlying models are good. Very few products are pleasant to delegate to.

Distribution through existing channels keeps proving out. Poke's traction came from living where people already type. Cognition put the figure at more than 100 million messages exchanged in the three months before the acquisition. No app to install, no new habit, no dashboard — the same bet we made with iMessage, Slack, Teams, and the phone.

Consumer and work assistants are separating. Poke was built for one person's life. Devin is a tool for engineers. Merging the personality of the first into the second is a coherent strategy, but it is not a strategy about team workflows, approvals, or audit. That gap is the category question of the next year, and it is unrelated to how good any of these products are at their own job.

The questions this should add to your checklist

If you are evaluating any assistant right now — ours included — an acquisition in the category is a good prompt to ask the boring durability questions:

1. Who is the product for now? Not at launch. A consumer product inside an enterprise company, or vice versa, is not doomed, but its roadmap now has to win an internal argument it did not have to win before.
2. What happens to your data and integrations if the product pivots? Ask about export. Ask specifically what you get back and in what format.
3. Is the thing you're buying the model, or the interface? If it is the interface, that is a real asset — the deal just proved it. It is also the layer most easily rebuilt by whoever owns the distribution.
4. Does it fit how your team is accountable? Personality is what makes an assistant usable. Approvals and an audit trail are what make it deployable at a company. Products optimized purely for one person's convenience tend not to have the second, because a single user has no one to be accountable to.

Where we sit

Arlo is built for the team case: it runs across iMessage and SMS, Slack, Microsoft Teams, and live phone calls, holds sends and tool writes for a named reviewer, and logs every run. That is a different product from a personal texting assistant, and for one person's calendar and errands a personal assistant is often the better buy. We have said that on the Arlo vs Poke page since before this deal and it is still true.

If you were leaning toward Poke and the acquisition made you want to see the field again, the Poke alternatives roundup covers the options honestly, including where they beat us. For what an assistant costs to actually run once you're past the shortlist, we published our own per-task numbers.

Details in this post reflect public reporting as of August 2026 and will age. Check Cognition's and Poke's own announcements for the current state.

Frequently asked questions

Who acquired Poke? Cognition, the AI lab behind the Devin coding agent, acquired The Interaction Company — Poke's maker — in an announcement dated 23 July 2026, in a deal reported in the low nine figures.

Is Poke shutting down? No. Cognition said Poke continues to operate and gains its models and infrastructure. The stated rationale for the deal was to bring Poke's conversational style into Devin.

Why would a coding-agent company buy a texting assistant? For the interaction design. Cognition's stated reason was Poke's conversational style — evidence that in a market where the underlying models are broadly comparable, the interface and personality are what is scarce.

Should the acquisition change which AI assistant I buy? Not on its own. It is a reason to ask durability questions of any vendor: who the product is for now, what data export looks like, and whether the tool fits how your team approves and audits work.